Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[LSR] Fix matching vscale immediates #100080

Merged
merged 2 commits into from
Jul 24, 2024
Merged

[LSR] Fix matching vscale immediates #100080

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0fe5354
[LSR] Fix matching vscale immediates
MacDue Jul 23, 2024
69bb025
Review fixups
MacDue Jul 23, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 4 additions & 2 deletions llvm/lib/Transforms/Scalar/LoopStrengthReduce.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -946,13 +946,15 @@ static Immediate ExtractImmediate(const SCEV *&S, ScalarEvolution &SE) {
// FIXME: AR->getNoWrapFlags(SCEV::FlagNW)
SCEV::FlagAnyWrap);
return Result;
} else if (EnableVScaleImmediates)
if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(S))
} else if (const SCEVMulExpr *M = dyn_cast<SCEVMulExpr>(S)) {
if (EnableVScaleImmediates && M->getNumOperands() == 2) {
if (const SCEVConstant *C = dyn_cast<SCEVConstant>(M->getOperand(0)))
if (isa<SCEVVScale>(M->getOperand(1))) {
S = SE.getConstant(M->getType(), 0);
return Immediate::getScalable(C->getValue()->getSExtValue());
}
}
}
return Immediate::getZero();
}

Expand Down
20 changes: 12 additions & 8 deletions llvm/test/Transforms/LoopStrengthReduce/AArch64/vscale-fixups.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -384,27 +384,31 @@ for.exit:
ret void
}

;; This test demonstrates an incorrect MUL VL address calculation. Here there
;; are two writes that should be `16 * vscale * vscale` apart, however,
;; loop-strength-reduce has ignored the second `vscale` and offset the second
;; write by `#4, mul vl` which is an offset of `16 * vscale` dropping a vscale.
;; Here are two writes that should be `16 * vscale * vscale` apart, so MUL VL
;; addressing cannot be used to offset the second write, as for example,
;; `#4, mul vl` would only be an offset of `16 * vscale` (dropping a vscale).
define void @vscale_squared_offset(ptr %alloc) #0 {
; COMMON-LABEL: vscale_squared_offset:
; COMMON: // %bb.0: // %entry
; COMMON-NEXT: rdvl x9, #1
; COMMON-NEXT: fmov z0.s, #4.00000000
; COMMON-NEXT: mov x8, xzr
; COMMON-NEXT: cntw x9
; COMMON-NEXT: lsr x9, x9, #4
; COMMON-NEXT: fmov z1.s, #8.00000000
; COMMON-NEXT: cntw x10
; COMMON-NEXT: ptrue p0.s, vl1
; COMMON-NEXT: cmp x8, x9
; COMMON-NEXT: umull x9, w9, w9
; COMMON-NEXT: lsl x9, x9, #6
; COMMON-NEXT: cmp x8, x10
; COMMON-NEXT: b.ge .LBB6_2
; COMMON-NEXT: .LBB6_1: // %for.body
; COMMON-NEXT: // =>This Inner Loop Header: Depth=1
; COMMON-NEXT: add x11, x0, x9
; COMMON-NEXT: st1w { z0.s }, p0, [x0]
; COMMON-NEXT: add x8, x8, #1
; COMMON-NEXT: st1w { z1.s }, p0, [x0, #4, mul vl]
; COMMON-NEXT: st1w { z1.s }, p0, [x11]
; COMMON-NEXT: addvl x0, x0, #1
; COMMON-NEXT: cmp x8, x9
; COMMON-NEXT: cmp x8, x10
; COMMON-NEXT: b.lt .LBB6_1
; COMMON-NEXT: .LBB6_2: // %for.exit
; COMMON-NEXT: ret
Expand Down
Loading